Spike-and-slab type variable selection in the Cox proportional hazards model for high-dimensional features

Area of Science:

• Biostatistics
• Genomics
• Survival Analysis

Background:

• Variable selection is crucial in high-dimensional survival data analysis.
• Cox models are widely used for survival data but can face challenges with complex variable selection.
• Spike-and-slab priors offer a robust Bayesian approach for variable selection.

Purpose of the Study:

• To develop a novel variable selection framework for Cox models using spike-and-slab priors.
• To simplify the estimation equation in Cox models by transforming parameter spaces.
• To identify important genes associated with survival outcomes in lung adenocarcinoma.

Main Methods:

• Transformation of score and information functions for the partial likelihood.
• Utilizing the spike-and-slab prior distribution within a Gibbs sampling framework.
• Stochastic variable search for identifying covariate sparsity structures.
• Application to lung adenocarcinoma data for gene identification.

Main Results:

• The proposed method effectively performs variable selection in the context of Cox models.
• Numerical simulations demonstrate good finite-sample performance.
• Identification of potential key genes associated with decreased survival in lung adenocarcinoma.

Conclusions:

• The developed framework offers a stable and effective approach for variable selection in survival analysis.
• The method facilitates the identification of biologically relevant genes impacting disease prognosis.
• This approach enhances the analysis of high-dimensional genomic data in cancer research.